Irradiance vs Brute Force

[Morbid Angel]

Gi is just a bounce, a fill light if you will. In exterior its less noticeable bec its illuminated from all directions, in interior it would play a much more vital role as bounce light.

[super gnu]

yes a "multiframe incremental" for BF would be awesome. ive asked a number of times for variations of that or an "unbiased" mode for the imap. its sadly impossible to turn off interpolation for the imap, set the min and max to 0 and get a bf solution. its WAY slower tha BF and looks awful.

[glorybound]

I wouldn't image it worked for most exteriors. I realize that it is light bouncing around, a lot of the image would be dark.

Gi is just a bounce, a fill light if you will. In exterior its less noticeable bec its illuminated from all directions, in interior it would play a much more vital role as bounce light.

[Morbid Angel]

Here I've done some tests (I apologize in advance for any inconsistencies I have two kids bouncing on me as do this):

Its a free scene from the web. I like to use it because it has a variable level of complexity, carpet, and wall decor take a bit of time. I've used very simple settings:

base settings adaptive dmc 1/24, with 0.01, all unclamped with no subpixel, only max ray intensity (all defaults)

irmap/lc

lc is default all with 1200 samples

shade rate is 6

irradiance map medium preset with -3/-1 and 100/50 and clr thresh of 0.2

dmc 1/24 and 0.01

You can see irmap is pretty clean though there are edges of the ceiling which suffer from common irmap precision problems since its max rate is -1 and not 0.



http://postimg.org/image/xgtg7lqzx/full/

Second render is bf/lc



http://postimg.org/image/en7nalsrx/full/


You can see that the bf render is a bit more grainier and took 6 minutes where are the irmap is 4. Neither of them satisfy me to the fullest extent - irmap render needs higher samples for more accurate shadows, bf needs higher samples for less noise.

Both renders spent extra time in the detail areas, for irmap it took 2 minutes for prepass and spent lots of time in the carpet and decor. I also find that because irmap is interpolated the grain from lights stands out quite a bit, so you get clean wall with grainy light edges.

Next I took irmap and went from -1 to 0 and render time went up to 8 min 40 sec. This is mostly on prepass as the actual render was very fast. I can say that with this quality its acceptable for the edges and precision, I've also reduced the hsubdivs and interp samples to about 3rd for 0 res its applicable. I've set the noise thresh to 0.006



http://postimg.org/image/mii6pf2el/full/

I would say that I actually spent more time fiddling with irmap (tests) to get a good result then with bf so if you take into account several test renders I did to get it to look good it might offset the total render time for irmap frame.

For brute force I've just lowered the noise thresh since its the only way to get it cleaner to 0.006 and the render time was 8m 59 sec which is just a bit slower then irmap. But this solidifies what I said earlier.



http://postimg.org/image/6dvrw6z1p/full/


super gnu, since new vray's msr is pretty good with bf it also flies through the blank wall areas pretty fast.

Lets put this to real test, I've disabled the grey override.

Both renders for irmap and bf produced very similar render time.



http://postimg.org/image/dwex4tqel/full/




http://postimg.org/image/y8w46sv71/full/


My conclusion is that with irmap of -1 for max, you can get faster result but at a cost of precision. Otherwise the render time is the same.

[super gnu]

yes it is as ive observed in recent builds, but for me this just means imap needs some updating when imap was last worked on, it used to be many, many times faster than BF. and it should be.. its supposed to be doing much less work.

[3LP]

Interesting!

Would it be possible to give this scene a go?
Could you send the URL of where we can download this scene as you said it's a free one?

Cheers
Stan

[Morbid Angel]

Here is a google drive share, lmk if it doesn't work for ya.

https://drive.google.com/file/d/0B-S...ew?usp=sharing

[eyepiz]

My testing on a interior render at 4k this is what I got with equal quality.

Adaptive / Auto Subdivs
LC 1600
Noise Threshold of 0.005

IM/LC Detail Enhanced Custom IM settings: 1hr 2min
BF /LC:2hr 5min.

IM is typically two to three times faster than BF on most my projects.

[Morbid Angel]

yeah but I bet you didn't render the irmap at full res.

[eyepiz]

yeah but I bet you didn't render the irmap at full res.

Your right I didn't.. I see little to no benefit doing full res at 4k.

[Morbid Angel]

I would be curious to see your irmap render on grey shaded, to check precision of irmap.

[eyepiz]

I remembered seeing this somewhere..it's in the IM tool tip.

**Note that the presets are targeted for a typical 640x480 image. Larger images usually can do with lower Min/Max rates than those specified in the presets**

[glorybound]

Are we talking about turning off reflection, rendering out and saving the IRR map, to be used when you turn reflection back on? I use to do this, however, for some reason I stopped. Also, I have tried this with 1/2 resolution, but it didn't look good.

Bobby Parker www.bobby-parker.com

[glorybound]

Is there any debate that IRR is much faster than BF, or are some arguing that BF can be just as fast? I mean, if you spend longer playing with IRR settings than you do actually rendering, it might be an argument that BF is the way to go.

Bobby Parker www.bobby-parker.com

[eyepiz]

Is there any debate that IRR is much faster than BF, or are some arguing that BF can be just as fast? I mean, if you spend longer playing with IRR settings than you do actually rendering, it might be an argument that BF is the way to go.

Bobby Parker www.bobby-parker.com

I use almost same IRR settings with most my work, I rarely have to tweak them. The Image I compared I used 1/4 res on the IM map and I cant tell the difference from the BF one.